← Models

DeepSeek V4 Pro

DeepSeek · DeepSeek V4

DeepSeek's stronger V4 route for million-token reasoning, agentic coding, and higher-end open-weight workloads.

Part of DeepSeek V4 family · Other versions: DeepSeek V4 Flash
Type
language
Context
1M tokens
Max Output
384K tokens
Status
current
Input
$1.74/1M tok
Output
$3.48/1M tok
API Access
Yes
License
DeepSeek License
reasoning long-context agentic coding open-weights frontier
Released April 2026 · Updated May 16, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on May 16, 2026.

DeepSeek V4 Pro is the higher-capability route in the DeepSeek V4 family. DeepSeek describes it as a 1.6T-total-parameter, 49B-active-parameter model aimed at frontier-style reasoning, agentic coding, and long-context workloads.

The practical role is clear: use V4 Flash as the inexpensive default, then route harder reasoning, coding, and analysis tasks to V4 Pro when the quality lift is worth the extra cost.

Capabilities

DeepSeek positions V4 Pro around agentic coding, math, STEM reasoning, and broad world knowledge. It supports both thinking and non-thinking modes, which makes it more flexible than a dedicated reasoning-only endpoint.

The 1M-token context window makes it relevant for repository-scale coding assistance, long policy or contract analysis, large document bundles, and research workflows where a conventional 128K or 200K context window forces more retrieval choreography.

Technical Details

Official anchors at this snapshot:

  • Model ID: deepseek-v4-pro.
  • 1,000,000 token context length.
  • 384,000 maximum output tokens.
  • 1.6T total parameters and 49B active parameters in DeepSeek’s launch notes.
  • Thinking and non-thinking modes.
  • OpenAI Chat Completions and Anthropic-compatible API formats.
  • JSON output, tool calls, chat prefix completion, and FIM completion in non-thinking mode.

Pricing & Access

DeepSeek’s regular V4 Pro API pricing is listed per 1M tokens:

  • Cache-hit input: $0.0145
  • Cache-miss input: $1.74
  • Output: $3.48

DeepSeek is also running a temporary 75% discount through May 31, 2026 at 15:59 UTC, which lowers effective pricing to 0.003625cache−hitinput,0.003625 cache-hit input, 0.435 cache-miss input, and $0.87 output during the promo window. Production cost models should use the regular rates unless the workload is explicitly short-lived.

Best Use Cases

Choose V4 Pro for harder reasoning, complex coding, long-context research, and workflows where DeepSeek’s open-weight economics matter but V4 Flash is too shallow.

It is less ideal as a default high-volume assistant when V4 Flash gives enough quality, or when governance requirements point to a provider with clearer enterprise contracts.

Comparisons

  • DeepSeek V4 Flash: Cheaper and faster default; V4 Pro is the quality route.
  • Qwen3.6 Max Preview: Another strong Chinese frontier option; V4 Pro stands out on open-weight release plus 1M-context API framing.
  • GPT-5.5: Broader OpenAI ecosystem and product surface; V4 Pro offers much lower token pricing and open-weight deployment options.